{"id":"W4386242325","doi":"10.1145/3600211.3604690","title":"Self-Destructing Models: Increasing the Costs of Harmful Dual Uses of Foundation Models","year":2023,"lang":"en","type":"article","venue":"","topic":"Adversarial Robustness in Machine Learning","field":"Computer Science","cited_by":12,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"Open Philanthropy Project; Canadian Institute for Advanced Research","keywords":"Computer science; Adversarial system; Artificial intelligence; Machine learning; Reinforcement learning; Foundation (evidence); Dual (grammatical number)","routes":{"ca_aff":false,"ca_fund":true,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.01185027,0.001575391,0.001234668,0.0006488285,0.001088952,0.002485891,0.00360887,0.002353621,0.003139499],"category_scores_gemma":[0.05473512,0.001144468,0.001495587,0.0004932064,0.004686907,0.007984308,0.007242789,0.007533396,0.001291689],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0014697,"about_ca_system_score_gemma":0.00206812,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002158498,"about_ca_topic_score_gemma":0.002498203,"domain_scores_codex":[0.9938236,0.003173214,0.0002898886,0.0009320792,0.001245924,0.0005351317],"domain_scores_gemma":[0.9478292,0.02679013,0.002617107,0.02005914,0.001826144,0.000878345],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.0008468221,0.0003101738,0.012684,0.0002593137,0.0002385207,0.0005147411,0.000925694,0.6956021,0.01006216,0.1416006,0.006636214,0.1303196],"study_design_scores_gemma":[0.00003899165,0.0001261455,0.0003950354,0.00005690612,0.00004260812,0.000143738,0.00006904546,0.8979372,0.006900643,0.09114015,0.003113183,0.00003618809],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.1297434,0.0005170288,0.8564489,0.002614432,0.0001353956,0.0001599679,0.0002083332,0.004350345,0.00582218],"genre_scores_gemma":[0.8757085,0.0002705092,0.1184646,0.0008437575,0.00006853484,0.0002285419,0.0002719492,0.0009422521,0.003201263],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.01185027,"threshold_uncertainty_score":0.06267095,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.04462146224270767,"score_gpt":0.2854168640465077,"score_spread":0.2407954018038,"validation_status":"score_only:v0-immature-baseline","note":"Baseline scores from an immature model (maturity gate not passed). Scores rank; they never assert a category."}}